According to the classification of using multiple sensors, the obstacle detection and recognition technology of unmanned driving is summarized, and the main technical means, algorithms, advantages and disadvantages, difficulties and so on are discussed. These methods mainly include lidar and camera fusion, camera and radar fusion and lidar, camera and radar fusion. In addition, some research achievements in this field in recent years are introduced, and some fusion technologies used in this field are briefly analyzed and summarized.
Road target detection and recognition is of great significance in the current field of automatic driving, and in the road target detection and recognition, the
high precision of detection algorithm and fast
reasoning speed are very important for safe automatic driving. The YOLOv5 (YOLO, you only look once) target detection algorithm can be used to identify and analyze road targets on the way of vehicles, playing a
role in assisting driving and reducing safety risks. By
using the improved YOLOv5 target detection
algorithm to train the BDD100K dataset, the model obtained can significantly improve the recall rate and thus the accuracy. The improved YOLOv5 algorithm mainly uses K‐means algorithm to find the most appropriate anchors for data sets, and gets more accurate models through real‐time data augmentation training. The results show that, on the BDD100K test set, the MAP‐50 of the improved model can reach 51.8%, and compared with the performance of the original model, the improved model has significantly improved the target detection mAP. Compared with the manual model, the proposed model can detect the
target more accurately while guaranteeing the
detection speed.
In view of the problems of multi‐scale, densely distributed and difficult small target detection in real‐time road condition detection, a new improved
yolov5 real‐time target detection method was
proposed. The activation function in the original
spatial pyramid (SPPF) was replaced by ReLu in the
backbone network to enhance the expression ability of
the model. The bidirectional feature pyramid (BiFPN)
is used to replace the original feature pyramid network (FPN) + pixel aggregation network (PAN) in
the neck area. The bidirectional feature fusion
improves the utilization of multi‐scale semantic
features and strengthens the extraction of image deep
features. The convolutional attention mechanism
(CBAM) is introduced to further improve the feature extraction ability of the algorithm and make the algorithm pay more attention to useful information. The results of self‐made traffic condition data set show that the average accuracy of the improved yolov5 model reaches 81.3%, which is 10.9% higher than that of the original yolov5 model, showing better
detection accuracy and real‐time performance.
Compared with some mainstream target detection algorithms, this algorithm has certain advantages.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.